A Hybrid Conv-LSTM-Attention Framework for Short-Term PV Power Forecasting Incorporating Data from Neighboring Stations

• DOI: 10.20944/preprints202405.0318.v1
• Publication Date: 2024-05-07T08:23:10Z
• Authors: Feng Hu; Linghua Zhang; Jiaqi Wang

To enhance the safety of power grid operations, this study proposes a high-precision short-term photovoltaic prediction method that integrates information from surrounding pho-tovoltaic stations and Conv-LSTM-ATT model. In deep learning model, not only is numerical weather (NWP) data target station used as input features, but also highly correlated features nearby are incor-porated. The research begins by analyzing correlation between irradiance se-quences, along with distance factors, to calculate composite similarity index other regional stations. Stations high indices then selected sources. Subsequently, Bayesian optimization techniques employed find optimal fusion ratios. Ultimately, using data, mod-eling conducted via neural network. Experimental results con-firm superiority proposed which demonstrates higher predictive accuracy compared three classical models. strategy determined significantly enhances accuracy, reducing root mean square error (RMSE) test set 20.04%, 28.24%, 30.94% for types, respectively.